KPMG Survey Puts AI Agent Management on a Shrinking Human Team

A KPMG survey of 2,500 tech executives finds 92 percent expect managing AI agents to become an important skill within five years. The same staffing chart has digital assistants at 36 percent of core tech teams by 2027, while permanent human staff fall from 48 percent to 43 percent.

The Survey Behind the Skill Headline

KPMG International published Leading in the Intelligence Age: Excelling today, shaping tomorrow in January 2026, drawing on a survey of 2,500 tech executives in 27 countries. Fieldwork ran in 2025. Every firm in the sample books more than US$100 million in annual revenue. The regional split is 43 percent EMEA, 29 percent Asia-Pacific, and 28 percent the Americas.

88 percent of those organisations say they are already investing in building agentic AI into their systems. Guy Holland, global leader of KPMG’s CIO Center of Excellence, wrote that firms are moving past what he called the early phase of ‘AI roulette’, the habit of scattering bets across tools, and are trying to put AI into workflows and offerings.

Half of the leaders expect to reach top technology maturity in 2026. Only 11 percent say they are there now. Returns are thinner than the ambition. 74 percent say their AI use cases provide business value, yet only 24 percent say they are getting ROI across multiple use cases, a 7-percentage-point drop from KPMG’s prior survey.

THE SAMPLE IN BRIEF

• The panel: 2,500 tech executives at firms above US$100 million in revenue, across 27 countries.
• Agent spend: 88 percent say they are already putting agentic AI into systems, workflows, products, and value streams.
• The skill vote: 92 percent say managing AI agents will be an important skill within five years, a horizon the report’s 2026 publication dates to 2031.
• The maturity gap: 50 percent expect top tech maturity in 2026, against 11 percent who say they have it now.

KPMG’s India firm put the 88 and 92 percent findings on its official account in May 2026, months after the PDF landed.

𝗧𝗵𝗲 𝗻𝗲𝘄 𝗻𝗼𝗿𝗺𝗮𝗹 𝗶𝘀 𝗮𝗴𝗲𝗻𝘁𝗶𝗰. 88% of tech executives are already investing in agentic AI, and 92% say managing AI agents will be a core skill within five years.

To stay ahead, leaders need to prepare their workforce by building AI fluency and prioritising… pic.twitter.com/58MGJBJ0iP

— KPMG India (@KPMGIndia) May 19, 2026

Digital Assistants Take Eight Points of the Org Chart

Figure 8 in the report asks leaders to split their core tech team’s full-time equivalent capacity among three sources, now and in two years. Digital assistants, logged as the digital workforce on that chart, rise from 28 percent in 2025 to 36 percent in 2027. That is an eight-point jump. Permanent human staff fall from 48 percent to 43 percent. External contractors or consultants fall from 24 percent to 21 percent.

CORE TECH TEAM FTE MIX

The eight-point gain in digital assistants is funded by a five-point cut in permanent staff and a three-point cut in contractors. The report calls those human-side moves moderate. They still rewrite who sits on a core tech team. By 2027 the typical bench in this sample is expected to be 43 percent permanent humans, 36 percent digital assistants, and 21 percent contractors.

Holland’s foreword says most organisations anticipate only modest reductions in permanent human roles over the next two years. Five points off a 48 percent base is the modest reduction he is describing. The agents are not a side tool on that chart. They are the growing share of FTE.

A 5 Percent Club and a 4.5x Return

KPMG carves out a group it calls high performers, 5 percent of the executives surveyed. They score high on tech maturity, process maturity, and value. On value, the bar is a 200 percent or more return on digital-tech spend over the prior 12 months, counting revenue, cost savings, efficiency, productivity, and risk reduction, and excluding hoped-for future gains.

Those firms report an average ROI of 4.5x, more than double the 2x average across the full sample. They also plan to keep more people. High performers expect 50 percent of tech-team capacity to remain permanent human staff by 2027. Other organisations expect 42 percent. The overall Figure 8 number, 43 percent, sits between those two.

Zack Kass, a global AI advisor and former Head of Go-To-Market at OpenAI, is quoted in the report saying high performers expect about half their tech teams to be permanent human staff by 2027, a setup he describes as small, durable human cores running large AI-augmented groups. Umesh Sachdev, co-founder and CEO of Uniphore, tied the gap to leadership rather than tools.

Companies that learn to use AI and AI agents and all these architectures effectively are likely to leave their peer groups behind, and right now that is coming down to the leadership of companies and departments and teams.

Umesh Sachdev, co-founder and CEO, Uniphore, KPMG Global tech report 2026

The firms pulling away in this sample are not the ones emptying the human column fastest. They keep more permanent staff than the rest and still put agents on the team.

Ninety Percent Plan to Buy the Expertise

Just over half of the organisations, 53 percent, still lack the talent to bring digital transformation plans to life. The report says skill gaps that were tolerable in pilots get riskier at full scale. Against that shortage, 90 percent of tech executives plan to expand and strengthen their tech partnerships so they can reach expertise they do not have in-house.

Seth Patton, general manager of product marketing for Microsoft 365 Copilot, told KPMG that no organisation keeps up alone, and that this is true at Microsoft too. Noelle Russell, AI solutions architect, strategic advisor, and CEO of the AI Leadership Institute, argued for a split between what stays inside and what gets hired in.

Pick the areas that you want to keep in-house for domain expertise, then choose trusted partners to fill in the gaps across your portfolio.

Noelle Russell, AI solutions architect and strategic advisor, KPMG Global tech report 2026

KPMG is not only counting that demand. It sells into it. The firm’s own agentic practice page offers a TACO framework for classifying agents as Taskers, Automators, Collaborators, and Orchestrators, plus implementation services from strategy through workforce change. The survey’s 90 percent partnership plan is also a market the authors serve.

Nearly one-third of the tech executives plan to put more money into centres of excellence, which the report frames as a way to run cross-functional experiments under some control. That is still an internal build. The dominant move in the numbers is the 90 percent who intend to widen the partner net.

The Security Bill on Every New Partner

The same executives who plan to hire more partners already rank those relationships as risky. This year they put security concerns, and intellectual property and data-protection concerns, in their top five barriers to greater collaboration on emerging technologies.

WHAT THE PARTNER PUSH DRAGS IN

• Talent hole: 53 percent say they still lack the people to deliver their digital plans, which is why the partner channel is growing.
• Speed trade-offs: 69 percent of tech executives say programs chasing speed and lower cost cut corners on security, scalability, and data standardisation.
• Debt drag: 63 percent agree that the cost of fixing tech debt is holding back new work.
• Audit response: 36 percent of organisations plan to enhance data-sovereignty audits across their partnership ecosystems.

Russell’s follow-on line in the report is a discipline brief, not a slogan. “Paying attention to what you build means applying rigor and discipline to every model you select,” she said, pointing firms toward evaluation methods such as Stanford HAI’s HELM tests and the model leaderboards already wired into developer tools at Microsoft, Amazon, and NVIDIA.

The partnership story and the skill story collide here. Leaders vote managing agents a core skill, then plan to rent a large share of that skill, then list security and data protection as reasons those rentals are hard. The 36 percent who will tighten sovereignty audits are treating the partner layer as an attack surface, which it is once agents sit inside workflows and touch customer data.

An HR File for the Agents

Rohit Gupta, founder and CEO of Auditoria.AI, told KPMG that firms will need an HR-style database for AI agents, the way they already keep one for people. “We are already seeing job descriptions out there which say that the person will have to manage agents,” he said. Dean Bortz, director of AI go-to-market at Google, went further: every person will have to manage their own army of agents, including how those agents deal with each other and with the business, past the old CIO or CHRO boxes.

Kass’s operational advice is to shrink the human map. He wants smaller teams and flatter structures so the mix of people and agents can turn faster.

Play smaller, and you can be more forward-looking.

Zack Kass, global AI advisor and former Head of Go-To-Market, OpenAI, KPMG Global tech report 2026

That is already how the work looks in software groups that route issues turned into agent jobs instead of leaving tickets on a human backlog. The live picture of “managing agents” is a named roster of specialist bots under a thin human core, not a seminar on prompting. One human sets a task, a lead agent splits it, sub-agents finish the pieces, and the org chart fits on a screen. Kass’s smaller team is that screen.

Jenny Wood, group CIO at Skipton Group, asked the harder cultural question in the report: how to enable people to work a different way, rather than do the same work faster. Seth Patton put it as courage, whether a firm will change how it operates because of AI, not only how fast it ships tickets. The survey’s own FTE math is the change. Humans lose five points. Contractors lose three. Digital assistants take both.

Quantum Prep Trails the Agent Build-Out

While the sample crowds into agents, 41 percent of the executives say they are worried they are falling behind on threats from quantum computing and on putting post-quantum cryptography in place. The report notes that a working, stable quantum computer is still an open timing question, and that the near-term risk is breaking today’s encryption. Phil Mottram, executive vice president and chief sales officer at HPE, said people already worry about bad actors using quantum on the security side, and that HPE’s servers shipping now have been future-proofed against those threats.

The standards work is not theoretical. NIST released the first three finalized PQC standards in 2024, and the Secretary of Commerce approved FIPS 203, 204 and 205 on August 13, 2024. Dustin Moody, the NIST mathematician who heads the PQC project, has told organisations to start the transition immediately. In March 2025 NIST also selected HQC as a fifth algorithm, a backup for general encryption beside ML-KEM.

Seventy-eight percent of the KPMG sample agree they must take more risks on emerging technologies, including quantum and artificial superintelligence, to stay relevant. Risk appetite is high. Prep on the encryption the report itself flags is not. A 10-to-20-year history of folding new crypto into products is the reason NIST is pushing now, years before anyone can date a cryptographically relevant quantum machine. Harvest-now, decrypt-later collection of today’s traffic is already a reason to move.

Frequently Asked Questions

Who exactly answered the KPMG 2026 Global Tech Report?

Half of the 2,500 respondents are C-suite members, with 10 percent vice presidents, 31 percent directors, and 9 percent senior managers. They span eight industries, led by financial services at 31 percent, then technology and telecom at 11 percent, with consumer and retail, industrial manufacturing, healthcare and life sciences, energy, and automotive each at 10 percent, and government at 8 percent.

How does the report define a high performer?

High performers are 5 percent of the sample and must clear three bars at once: sit in the top two maturity stages in at least five of ten technology categories, sit at the highest process-maturity level in at least five of ten technology functions, and report a 200 percent or more return on digital-tech investment over the prior year. Their average reported return is 4.5x, against 2x for the full sample.

Which post-quantum cryptography standards can firms use now?

FIPS 203 specifies ML-KEM (from CRYSTALS-Kyber) for key establishment, FIPS 204 specifies ML-DSA (from CRYSTALS-Dilithium) for signatures, and FIPS 205 specifies SLH-DSA (from SPHINCS+) as a hash-based signature scheme. NIST later selected HQC in March 2025 as a backup encryption algorithm that uses different math from ML-KEM, and a FALCON-based signature standard has been queued as FIPS 206.

When did KPMG collect the Global Tech Report 2026 data?

The report states that research was conducted in 2025, then published in January 2026, with the view stretched to predictions for 2026 and beyond because the authors judged the tech cycle too fast for a backward-looking snapshot. The 2025 versus 2027 team mix is therefore a forecast from 2025 fieldwork, not a 2026 recensus.

HPE’s Mottram can already point to servers on the loading dock that have been future-proofed against quantum threats. 41 percent of the executives in KPMG’s sample still worry they have not done that work. The agent seats fill by 2027. The encryption standards they need have been on a public list since August 2024.